An artificial “brain” built by a 17-year-old whiz kid from Florida is able to accurately assess tissue samples for signs of breast cancer, providing more confidence to a minimally invasive procedure.
The cloud-based neural network took top prize in this year’s Google Science Fair.
Her first neural network played soccer.
For her Google Science Fair project, she built a neural network with Java and then deployed it to the cloud. She ran 7.6 million trials on it and found it is 99.1 percent sensitive to malignancy.
“As I get more data, the success rate will go up and the inconclusive rate will go down,” she said. “So with more data, I think it is hospital ready.”
She also compared her custom-built network to three off-the-shelf products that she tweaked to work with her dataset. Her sensitivity to malignancy was 4.97 percent better than the commercial networks.
Now that her network is built and shown to get smarter with experience, Wenger aims to deploy it in hospitals. She also wants to extend it to other types of cancer.
“It will require a little bit of coding and tweaking, but it would be very easy to adapt it so it could diagnose other types of cancer and potentially other medical problems,” she said.
- More Here
The cloud-based neural network took top prize in this year’s Google Science Fair.
Her first neural network played soccer.
For her Google Science Fair project, she built a neural network with Java and then deployed it to the cloud. She ran 7.6 million trials on it and found it is 99.1 percent sensitive to malignancy.
“As I get more data, the success rate will go up and the inconclusive rate will go down,” she said. “So with more data, I think it is hospital ready.”
She also compared her custom-built network to three off-the-shelf products that she tweaked to work with her dataset. Her sensitivity to malignancy was 4.97 percent better than the commercial networks.
Now that her network is built and shown to get smarter with experience, Wenger aims to deploy it in hospitals. She also wants to extend it to other types of cancer.
“It will require a little bit of coding and tweaking, but it would be very easy to adapt it so it could diagnose other types of cancer and potentially other medical problems,” she said.
- More Here
No comments:
Post a Comment